Even after a career spent at the forefront of AI research and development, I can say with confidence that we are in an unprecedented moment. While AI is undoubtedly the most disruptive technology of our generation, in the world of cybersecurity, hype doesn’t stop threats. Turning the immense promise of generative AI, deep learning and machine learning into tangible security outcomes requires more than just access to new models; it demands a disciplined and purposeful research philosophy.
Our philosophy rejects the traditional, isolated “ivory tower” of academic research and corporate R&D. Instead, it is relentlessly focused on real-world outcomes that are deeply embedded within the teams building the products, secure by its design and openly collaborative. This blueprint guides our work and is built upon four core principles that I believe are essential for making AI work for security.
1. Research in the Trenches
This philosophy begins with a foundational decision about structure. Instead of isolating researchers, we embed them directly within our product organizations for a simple reason: In cybersecurity, proximity to the place where security problems are solved is everything.
This structure ensures our researchers are focused on solving real-world problems, not abstract ones. Researchers sit alongside the engineers who build our products and the product managers who live and breathe our customers’ challenges. This proximity makes the transfer of technology seamless and organic. It fosters a constant dialogue that ensures our long-term, high-risk research projects remain grounded in what will ultimately make our customers safer.
2. Better Security Outcomes: The Only Metric That Matters
Many research organizations measure success by the number of academic papers they publish. We don’t. Our primary metric is the tangible improvement our research delivers to our products and, by extension, to our customers’ security.
This focus has two critical implications. First, it means that we train and evaluate our models on real-world security system data, not on sanitized “toy problems.” This ensures our AI is effective in the complex, messy reality of a live security environment. Second, it gives our teams the freedom to fail. We encourage a “fail-fast” mentality, enabling us to quickly discard ideas that don’t show promise and double down on those that do, without the pressure of a publication quota. Our goal is to build a portfolio of proven, effective AI for security, not to build a library of papers.
3. Security for AI, Not Just AI for Security
As the leader in cybersecurity, we have a dual responsibility: to build AI that promotes security, and to ensure the AI we — and our customers — build is itself secure. Our customers entrust us with their most sensitive system data, and protecting it is our highest priority.
This principle extends to our entire research operation. Our models are developed in highly secure environments, protected by our own best-in-class security products and secure-by-design frameworks. We meticulously vet our AI security and protect against model theft, prompt injections and other emerging threats. This may seem obvious, but secure AI is impossible without first having a strong understanding of AI security. Our commitment to this principle extends beyond our own walls; it is the core of our promise to our customers. The same security platform that protects our research is the one we offer to the industry, ensuring everyone can benefit from the lessons we learn on the frontlines of AI innovation.
4. Create a Community, Not a Fortress
Finally, we believe the best ideas come from collaboration. We actively avoid a “not invented here” mentality. Our teams are empowered to leverage the best innovations from the broader research community. These innovations include using LLM coding agents to make our own researchers more productive and generating synthetic data to make our models more robust.
We are committed to being active participants in the global conversation. We encourage our researchers to attend conferences, organize workshops, and continuously learn from what others have done. Progress in this field is a collective effort, and our goal is to contribute to and benefit from the shared knowledge of the entire ecosystem.
A Safer, Secure Future
Ultimately, these principles create a research engine that is both innovative and accountable. It’s an approach designed to turn cutting-edge science into real-world security, ensuring that the power of Precision AI delivers on its most important promise: a safer digital future for everyone.
To see these principles in action and explore our deep research into securing this next wave, read our full paper, Achieving a Secure AI Agent Ecosystem.
